Objective We set out to test the discriminative power of an age-adjusted upper reference limit for cerebrospinal fluid total protein (CSF-TP) in identifying clinically relevant causes of albuminocytological dissociation (ACD).
Methods We reviewed the charts of 2627 patients who underwent a lumbar puncture at a tertiary care centre over a 20-year period. Samples with CSF-TP above 45 mg/dL (0.45 g/L) were included. Samples with white cell count >5×109/L, red cell count >50×109/L and glucose <2.5 mmol/L (45 mg/dL) were excluded as were samples with incomplete data and those taken from paediatric patients (ie, age <18 years old). Patients with CSF-TP elevated above 45 mg/dL were considered to have ‘pseudo’ ACD unless their CSF-TP was in excess of age-adjusted norms in which case they were considered to have ‘true’ ACD. Adjustment for sex was not applied to the age-adjusted norms although the importance of gender has been previously described.
Results The presence of ACD was associated with a broad range of neurological diagnoses. Among all 2627 patients with ACD, a clinical diagnosis explaining CSF-TP elevation was identified in 57% of cases. ‘True’ ACD was associated with a suitable diagnosis in 75% of cases, whereas patients with ‘pseudo’ ACD showed an appropriate diagnosis in only 51% of cases. Use of an age-adjusted upper reference limit favoured the detection of polyneuropathy patients (13.5% proportionate increase) and excluded a larger number of patients with isolated headache (10.7% proportionate decrease; p<0.0001).
Conclusions Elevated CSF-TP is a common finding, with a range of underlying causes. Use of an age-adjusted upper reference limit for the CSF-TP value improves diagnostic specificity and helps to avoid overdiagnosis of ACD.
- neuromuscular disease
- adult neurology
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
Strengths and limitations of this study
This study is of the largest of its kind, incorporating the results of 2627 retrospectively collected cerebrospinal fluid (CSF) samples over 20 years.
Charts were thoroughly evaluated for potential sources of albuminocytological dissociation (ACD) including the review consultation notes, neuroimaging and laboratory data.
Sample size was sufficient to show a clear shift towards clinically relevant diagnoses like polyneuropathy at the exclusion of those without such pathology (eg, benign headache) using age-adjusted CSF-total protein (TP) thresholds in place of a traditional fixed threshold (ie, CSF-TP ≥0.45 g/L).
Charts of those patients with CSF samples demonstrating ACD (2627 of 16 045) were included using strict criteria (white cell count ≤5×106/L, red cell count ≤50×106/L, glucose ≥2.5 mmol/L, age ≥18 years, CSF-TP ≥0.45 g/L).
Factors previously described to vary with CSF-TP such as sex, body mass index, CSF sample number or lifestyle factors (smoking, alcohol or physical activity) were not explored as much of these data were unavailable at the time of analysis.
The term ‘albuminocytological dissociation’ (ACD) was first coined by Sicard and Foix in 1912 to describe the unexpected finding of elevated cerebrospinal fluid (CSF) protein without pleocytosis in patients with spinal compression.1 Four years later, the term became entrenched in the medical literature with the landmark article of Guillain, Barré and Strohl, describing the acute demyelinating polyradiculoneuropathy that now carries their name.2
We recently published CSF total protein (CSF-TP) reference intervals derived from institutional data at the Ottawa Hospital, comprising an initial dataset of 19 591 CSF samples analysed over a period of 20 years.3 After exclusions based on laboratory parameters (white cell count (WCC) >5×109/L, red cell count (RCC) >50×109/L and glucose <2.5 mmol/L) and 60 conditions associated with elevated CSF-TP, we determined age-adjusted continuous reference intervals and suggested that these would be more accurate than a commonly employed cut-off of 0.45 g/L (45 mg/dL).
In the current study, we hypothesised that the implementation of age-adjusted upper reference limits (URLs) would result in a larger proportion of identified patients with expectedly high CSF-TP protein—including those with inflammatory neuropathies. We, therefore, sought to describe the types of clinical diagnoses associated with ACD. Our aim was to distinguish between patients with ‘traditional’ ACD (CSF-TP >0.45 g/L), ‘true’ ACD (CSF-TP >age-adjusted reference limit) and those with ‘pseudo’ ACD (0.45 g/L<CSF TP<age-adjusted reference limit)—and to compare the types and frequencies of clinical diagnoses in each group.
Registrations and study population
All data were extracted from the Ottawa Hospital Data Warehouse based on CSF samples collected between 1 January 1996 and 31 December 2016. Laboratory data obtained from the database included CSF-TP, CSF glucose, CSF WCC, CSF RCC, and serum creatinine and TP results. In addition, demographics (age and sex) and clinical diagnostic codes (ICD-9/10 codes) were recorded. To identify the subset of patients with ACD, we applied specific inclusion/exclusion criteria (figure 1). Excluded were samples with CSF constituents otherwise outside of established thresholds including WCC >5×109/L, RCC >50×109/L, glucose <2.5 mmol/L (45 mg/dL) and CSF-TP <0.45 g/L (<45 mg/dL). Samples with incomplete clinical or laboratory data, or those performed on paediatric patients were also excluded.
Patient and public involvement
The study analysis used anonymised patient data extracted from the Ottawa Hospital Data Warehouse as described above.
Review of our database revealed that diagnostic codes (ICD-9/10 codes) generated at the time of lumbar puncture (LP) did not always reflect the ultimate diagnostic outcome. To ensure accuracy and quality of data, all 2627 clinical charts were reviewed with the goals of identifying: (1) the presence of any clinical condition known or suspected to cause increased CSF-TP and (2) the indication for performing a LP. The reference list of medical conditions believed to be associated with elevated CSF-TP was established based on a thorough search of the medical literature (table 1). Where the literature was unclear as to an expectation of CSF-TP elevation, a consensus was reached between the reviewers (JAB, PF and PRB). Cases in which multiple factors may have contributed to increased CSF-TP were discussed between reviewers (JAB, PF and PRB), to ensure the accuracy of classification. Each patient was subsequently categorised based on the most likely cause of high CSF-TP; if a cause was not found, patients were categorised based on the clinical indication for LP.
Technical specifications for the analytical equipment used in CSF analysis have been outlined in the methods section of the 2017 manuscript by McCudden et al 3:
CSF-TP was analysed on three different instruments over the course of the 20 years included in the study as follows: Roche Hitachi 917, 1 January 1996 to 30 September 2001; Beckman Lx20, 30 September 2001 to 1 April 2009 and Siemens Vista 1500, 1 April 2009 to 1 December 2016. The Roche method is based on a benzethonium chloride turbidimetric analysis, whereas the Beckman and Siemens methods use a pyrogallol red-molybdate complex, which is measured at 600 nm. In all cases, analyses were performed according to the manufacturer’s directions.
Other laboratory values were measured on different instruments across the two decades included in the study. Serum creatinine, TP and CSF glucose were measured on the platforms described above for the same time frames; creatinine was measured by the Jaffe method from 30 September 2001 to 13 April 2013 and by the enzymatic method thereafter. CSF WCC and RCC counts were determined using the Beckman Coulter between 1996 and 2009 and the Sysmex XE5000 from 2009 to 2016. All laboratory analyses were determined according to the manufacturer’s instructions throughout the study in a routine clinical laboratory in an academic medical centre (The Ottawa Hospital).
ACD was defined as ‘traditional’ ACD if CSF-TP exceeded a typical cut-off of 0.45 g/L. ACD was defined as ‘true’ ACD if the CSF-TP exceeded age-adjusted reference limits, as defined in McCudden et al. Age-adjusted reference limits were computed using the following formula:
If CSF-TP was between 0.45 g/L and the age-adjusted reference limits, the case was labelled ‘pseudo’ ACD. Based on the clinical diagnoses/categories and our review of the medical literature, patients were also divided into those with an expected increase in CSF-TP (patients possessing explanatory underlying conditions), and those where an increase was unexpected (a patient possessing no explanatory condition). For the proportion of patients with ‘true’ versus ‘pseudo’ ACD, the expectation and CIs were derived using bootstrap analysis given that all patients where ACD was unexpected were part of the original dataset from which the age-adjusted limits were derived.
For each clinical category, the category’s share of ACD patients was computed (ie, the number of patients assigned to a clinical category vs the remainder assigned to all other clinical categories). These proportions were compared for ‘traditional’ ACD and ‘true’ ACD using a Fischer’s exact test (table 2). Within the ‘traditional’, ‘true’ and ‘pseudo’ ACD groups, the frequencies of the underlying clinical categories were plotted in bar graph format for illustration (figure 2). Given the large number of statistical tests performed (ie, 47 Fischer’s exact tests), Bonferroni correction was applied to the threshold p value suggesting statistical significance (ie, 0.0001=0.05/47).
Given the established utility of ACD in polyneuropathy, a subgroup analysis focused on these patients. The median CSF-TP levels, shown with their IQR, were computed and compared with available literature. A Fisher’s exact test was also used to compare the relative shifts in ACD classification between inflammatory and non-inflammatory neuropathies when using the ‘traditional’ versus ‘true’ definition of ACD. Furthermore, a Mann-Whitney-Wilcoxon test was used to compare the mean CSF-TP of inflammatory and non-inflammatory neuropathies.
All statistical calculations and graphs were generated using R V.3.3.3 (The R Foundation, Vienna, Austria).
The range of CSF values among 2627 patients (1093 female with a median (IQR) age of 54 (25)) with ACD over a 20-year time frame has been plotted in figure 3. Among all patients with ‘traditional’ ACD (CSF-TP >0.45 g/L), the underlying clinical category/diagnosis was considered sufficiently explanatory in 56% (53%, 59%) (1474/2627) of cases. The finding of ‘true’ ACD was expected in 75% (0.72, 0.78) (446/597) of cases; whereas in ‘pseudo’ ACD, ACD was expected in only 51% (48%, 54%) (1028/2030) of cases (p<0.001). The relative number of cases identified and their specific diagnosis are shown in figure 2.
Table 2 lists the clinical categories/diagnoses used to classify patients and demonstrates the effect of using a data-driven age-adjusted reference limit as opposed to a traditional 0.45 g/L on the proportion of patients demonstrating ACD. Where CSF-TP elevation was unexpected, applying age-adjusted reference limits either decreased or did not change the proportion of these patients relative to other clinical categories. The opposite was true for the clinical categories where CSF-TP elevation was expected, as a significant increase in the relative share of patients with ACD was seen or there was no significant change. A notable exception to this pattern was inflammatory white matter disease where a large but non-significant decrease in the relative share of patients who would have been traditionally classified with ACD was seen.
When the subgroup of patients with polyneuropathy was examined, the effect of applying an age-adjusted URL tended to be more pronounced in those patients with non-inflammatory neuropathies (although p=0.25, thus, statistical significance was not reached). When the age-adjusted URL was applied, the number of non-inflammatory neuropathy patients exhibiting ACD decreased by 65% (from 17 to 6), in contrast, those with inflammatory neuropathy showed only a 35% reduction in cases (from 187 to 121). Moreover, the mean CSF-TP showed significant difference, measuring 1.05 g/L (0.85 g/L) for inflammatory neuropathy versus 0.57 g/L (0.16 g/L) for non-inflammatory neuropathy (p<0.001).
ACD has been described in a large number of peripheral and central nervous system disorders. Several disease-specific mechanisms have been proposed, including: (1) the intrathecal production or liberation of proteins such as IgG and myelin basic protein, (2) blood–brain barrier dysfunction in meningeal or parameningeal inflammation, (3) blood–nerve barrier dysfunction in neuropathy, (4) sequestration of CSF in spinal compression or (5) decreased CSF flow. Minor elevations of CSF-TP that are not associated with increased cell counts have also been linked to various attributes. This would include differences due to sex, age, body mass index and maximal abdominal circumference.4 Techniques have been proposed to correct for the impact of physiological variables, such as age, on metrics of blood CSF barrier dysfunction including those more tailored towards such an assessment (eg, albumin quotient).5 The result of routine CSF testing however often still leaves clinicians with a need to decide what level of isolated protein elevation may reflect an abnormality requiring further investigation.
ACD was a remarkably common CSF finding in diagnostic LP at our institution, present in 2627 of 8340 specimens (or 31.4%) using the traditional 0.45 g/L reference limit. This was proportion was similar to that observed in a publication by Hegen et al where CSF TP elevation was present in 31.8% of samples.6 We found however that ACD was only present in 597 (or 7%) with age-adjusted institutional reference limits. Of those patients with ‘true’ ACD, the most frequently associated clinical diagnoses were polyneuropathy (21%), benign headache (14%), seizures (9%) and intra-axial/extra-axial tumours (8%). There was, therefore, a marked reduction in the number of patients meeting criteria for ACD particularly in patients with clinical diagnoses not expected to be associated with ACD (benign headaches, transient encephalopathy and others), who often exhibited ‘pseudo’ ACD. Conversely, reductions in ACD frequency were less prominent in diagnostic categories where ACD has been well described, such as inflammatory polyneuropathy. Moreover, in those patients with ‘true’ ACD, the underlying clinical diagnosis was considered to be the potential cause of the protein elevation in 75% (72%, 78%) of cases.
Brettschneider et al 7 similarly observed frequencies of particular clinical diagnoses (resulting in higher specificity for apparently causal conditions), when age-adjusted reference limits were applied, though their study used the serum albumin quotient (Qalb) rather than CSF-TP. Similar to our findings, they observed that in patients with what we qualify in our report as ‘true’ ACD, 73% had an explanatory cause of Qalb elevation (including Guillian-Barré syndrome (GBS)/Chronic Inflammatory Demyelinating Polyneuropathy (CIDP), lumbar spinal stenosis or epileptic seizures, among other diagnoses). While the Brettschneider et al article aligns with our findings, their sample size was significantly smaller (only 367 patients with ACD were studied).
In our dataset and the Brettschneider et al study, patients with polyneuropathy were found to be the main source of clinically relevant (expected) ACD. Many articles in the medical literature have focused on the detection of ACD in polyneuropathy, for the purpose of identifying those patients with immune/demyelinating neuropathies. In inflammatory neuropathies (including GBS and CIDP), ACD is considered one of the cardinal diagnostic features, with mean CSF-TP levels in excess of 1.0 g/L (100 mg/dL) in some reports.8 9 Non-inflammatory neuropathies often display a more modest degree of blood–nerve barrier dysfunction as evidenced by less extreme elevations in CSF-TP.9 10 Providing CSF-TP thresholds that consider age adjustment may explain some element of the mild elevation seen in non-inflammatory neuropathy and therefore aid in distinguishing them from their inflammatory counterparts. The significance of this has been highlighted in the study by Allen et al which examined the diagnosis and misdiagnosis of CIDP in 59 consecutive patients. They showed that over-reliance on mild elevations of CSF-TP was often a source of false CIDP diagnoses. Moreover, they showed that once reclassified using European Federation of Neurological Societies criteria, patients with CIDP had a substantially higher mean CSF-TP (1.56 g/L) as compared with those without CIDP (0.61 g/L). To put this roughly into the context of our previously derived population norms, the median age of those falsely diagnosed with CIDP was 49.8 years for which our estimates suggest 0.59 g/L (59 mg/dL) as a more appropriate threshold for the CSF-TP URL (ie, the computed estimate of the 97.5th percentile) than a more traditional 0.45 mg/dL.3 This paper by Allen et al, therefore, underscores the need to explore techniques like age-adjusting CSF-TP URLs as a potential means to reduce misdiagnosis of CIDP.
Other notable clinical categories included a headache and inflammatory white matter disease. From examining the data, one may question why a benign headache might be so prominently represented in a sample of patients with ACD. We suspect that this reflects the volume of patients who underwent LP as screening for subarachnoid haemorrhage or meningitis to investigate a common and non-specific symptom, namely headache, in the context of an overly sensitive age-invariant threshold. To that point, headache patients were the most likely to be reclassified as ‘pseudo’ ACD when age-adjusted thresholds were applied. Similarly, patients with inflammatory white matter showed a high likelihood of being reclassified as ‘pseudo’ ACD when age-adjusted thresholds were applied. We suspect that this relates to the mild degree of CSF-TP elevation noted in multiple sclerosis—likely as a result of less aggressive and more chronic blood–brain barrier dysfunction.11 12
Our study does have several limitations worth mention. First, without a formal chart review of all 16 045 patients with complete laboratory data (especially those with CSF-TP <0.45), we are unable to formally quantify the sensitivity and specificity of CSF analysis for particular diagnoses. Second, we did not take into account the effect of sex, body mass index, CSF sample number or lifestyle factors (smoking, alcohol or physical activity) on CSF-TP levels.4 Third, we believe that a proportion of CSF-TP variability remains unexplained and prospective data collection (including additional laboratory values such as glycosylated haemoglobin and thyroid stimulating hormone) may further improve our understanding of CSF-TP variability.13 Fourth, three different instruments were used to measure CSF-TP over the course of the study. Although 95% CIs for age-and-instrument-partitioned intervals overlapped for ages <65 years; for >65 years, a modest but statistically significant difference in CSF-TP was found between devices as outlined in our previous paper. This raises the importance of device calibration and the potential impact on the interpretation of borderline CSF-TP levels.3 Fifth, although our median estimates of CSF-TP in inflammatory and non-inflammatory neuropathy appear to align with previously reported values, they represent a biased sample where those with CSF-TP <0.45 were excluded. Sixth, out of 19 591 samples, only 2627 samples were included in the analysis after eliminating repeat and incomplete sampling as well as those with biochemical and cytological measures outside of established norms and thus are not generalisable to those patients who have additional CSF abnormalities (eg, pleocytosis, hypoglycorrhachia or high RCC).
Apart from the above limitations, we believe that our study successfully presents the relevant clinical diagnoses associated with ‘true’ ACD, above the age-adjusted URL. In addition, our analysis highlights that the use of age-adjusted CSF-TP thresholds seems to increase the specificity for clinically relevant (expected) conditions. We would, however, caution clinicians not to overemphasise the importance of a finding of ACD, particularly given that common conditions such as lumbar stenosis may be the cause. To maximise the insight gained from CSF-TP levels, a future study evaluating the effects of additional factors on values within the ‘true’ ACD range is warranted.
Priya Figurado (Department of Medicine, Division of Neurology, University of Ottawa, Ottawa, Canada) who performed data collection; Michel Shamy (Department of Medicine, Division of Neurology, University of Ottawa, Ottawa, Canada; The Ottawa Hospital Research Institute, Ottawa, Canada) who performed substantive editing.
Patient consent for publication Not required.
Contributors Conceptualisation, methodology: CM, PRB and JAB. Writing and revision: JAB, AB, PRB and CM. Supervision: PRB and CM. JAB was involved with conceptualisation, writing and revision of the current study. CM was involved with conceptualisation, writing and revision of the current study and supervision as it pertained to the current study. AB writing and revision of the current study. PRB was involved with conceptualisation, writing and revision of the current study and supervision as it pertained to the current study.
Funding JAB currently receives fellowship funding from the Canadian Network of MS Clinics which is supported through contributions from Biogen Idec Canada, EMD Serono, Sanofi Genzyme, Novartis Pharmaceuticals Canada., Hoffmann-La Roche and Teva Canada Innovation. AB reports grant funding from the GBS-CIDP Foundation International and Grifols.
Disclaimer This study did not receive any direct financial support. CM reports no disclosures. PRB reports no disclosures.
Competing interests None declared.
Ethics approval This study was approved by the Ottawa Hospital Research Institute (OHRI) Ethics Board (protocol #20160863–01H).
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Additional data regarding disease specific cerebrospinal total protein levels can be accessed by emailing John Brooks at John.firstname.lastname@example.org.
Author note Albuminocytologic dissociation is seen in a broad range of neurological diagnoses. Our study of 2627 patients with ACD examines the causes.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.